Drop unused county hospitalization aheads early in pipeline instead of in dashboard

Hospitalizations forecasts include predictions for 1-28 days ahead, versus the 1-4 weeks ahead used for cases and deaths. However, in the dashboard we actually display only 4 of the 28 aheads for hospitalizations, so we waste a lot of time in the pipeline loading and scoring, and in the [dashboard filtering](https://github.com/cmu-delphi/forecast-eval/blob/b81379936d2e88bf37f90a9fd3100d4dbe3888ad/app/R/data_manipulation.R#L58) those extra aheads.

The extra aheads were originally included in the pipeline as a "just in case" but haven't been useful yet.

Start in app/R/data_manipulation.R at the linked dashboard filtering code, then trace where hospitalization aheads are loaded and scored earlier in the pipeline. Done means the unused hospitalization aheads are removed before those stages while the dashboard still displays the intended four aheads.

Written by the indexing model from the issue text.